DEVELOPMENT OF BATTERY STATE OF CHARGE ESTIMATION BASED ON DATA-DRIVEN USING DEEP LEARNING METHOD

• Main Author: Wily, Farhan
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73128
• Institution: Institut Teknologi Bandung

Accurate State of Charge (SOC) estimation is a crucial parameter for Battery Management Systems (BMS) to monitor and prevent batteries from experiencing overcharge and overdischarge, which can degrade battery performance and lifespan. Direct measurement-based and model-based methods can provide good accuracy in SOC estimation, but they heavily rely on environmental conditions and require specific testing procedures for each battery model used, making them challenging to implement in real-world systems. Data-driven methods have gained significant attention as they are not dependent on environmental conditions or battery models. These methods utilize large amounts of historical data, making deep learning models well-suited for handling such data with high accuracy. In this study, we employed Deep Neural Network (DNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM) algorithms, which are deep learning algorithms, for SOC estimation. This research was conducted on a timeseries dataset of lithium NCA battery cycles. The model development process followed the Cross Industry Standard Process for Data Mining (CRISP-DM) framework. In some cases, the features present in the raw dataset did not yield the best performance. In such cases, an analysis was performed on the input variables used in SOC estimation, and it was found that incorporating the Open Circuit Voltage (OCV) and battery energy as input variables allowed the model to be more sensitive to SOC changes and maximize estimation accuracy. Experimental results indicated that the RNN algorithm outperformed DNN and LSTM based on performance metrics. SOC estimation using the RNN method on dynamic battery cycle test data achieved R2 of 0.98581, RMSE of 2.49243%, MAE of 1.97653%, and MAPE of 4.08%. These findings and results provide valuable insights for the development of improved BMS to enhance battery performance and lifespan in various applications.